Run Claude Code, Codex, Cursor, Aider or your own app on 350+ free models from 66 providers. When one free plan runs out, the router switches to the next one — instantly.
curl -fsSL https://free-llm.com/install.sh | sh
macOS · Linux · Windows, npm & Docker · needs Node.js 18+
Dozens of companies give AI away for free. But each free plan is small — a few requests per minute, a daily cap — and each has its own key, own address and own rules. When one runs out in the middle of your work, your tool simply stops.
Claude Code, Cursor or your script sends a normal OpenAI- or Anthropic-style request to localhost:4000.
The best one that is up right now — by quality, speed and quota left. It learns from every answer.
Rate limit, daily quota, error or silence: it switches to another free model before your app notices.
Claude Code · Codex · Cursor · your app │ one address, one key ▼ ⚡ Free-LLM Router (on your computer) ┌──────┬──────┬────────┬──────────┬────────┐ ▼ ▼ ▼ ▼ ▼ ▼ Gemini Groq Mistral OpenRouter NVIDIA …61 more
The big advantage
Many small free plans, stacked together, behave like one big free plan — that doesn't go down.
66
free providers it can use
350+
free models
25
need no credit card
$0
no premium tier, ever
Numbers are live from the free-llm.com directory. New free model this morning? It reaches every router the same day.
It starts a short guided setup: it shows the best free providers, you paste a key or two, it tests them, and it can configure your coding tools for you. Then open http://localhost:4000.
curl -fsSL https://free-llm.com/install.sh | sh
No admin rights, nothing global: installs into ~/.free-llm and adds a `free-llm` command.
🚀 Works with zero keys
Guest mode answers right away (slow, limited) so you can see it work before signing up anywhere.
🔑 Then add 2–3 free keys
Google AI Studio, Groq, OpenRouter… about a minute each, most without a credit card.
🖥️ Dashboard included
Add and test keys, see which models are ready, chat playground, live activity, settings.
Coding agents are configured with one command (your previous settings are backed up, --remove undoes it). Everything else gets a copy-paste recipe in the dashboard.
free-llm setup claude
free-llm setup codex
free-llm setup opencode
free-llm setup aider
free-llm launch qwen
free-llm launch goose
free-llm launch crush
Settings → Models → base URL
OpenAI Compatible provider
OpenAI Compatible provider
OpenAI Compatible provider
config.yaml (chat + autocomplete)
Ollama or OpenAI-compatible
Ollama / OpenAI-compatible
Connections → OpenAI API
custom endpoint
OpenAI provider
base_url = …/v1
base_url = router
ChatOpenAI(base_url=…)
createOpenAI({ baseURL })
free models as a tool
Cursor calls custom endpoints from its own servers: run the router on a server, or expose it with a tunnel. All other tools work with the router on your own computer.
Send one of these as the model name — or any model id like deepseek-v4-flash, and the router finds a provider that serves it for free.
| auto | Best overall for each request |
| auto:coding | Strongest coding models first — for agents |
| auto:fast | Lowest latency first — chat, autocomplete |
| auto:smart | Most capable models first |
| auto:long | Largest context windows first |
| auto:vision | Models that can read images |
| auto:reasoning | Thinking / reasoning models first |
| auto:spread | Rotate providers to stretch free quotas |
| auto:fusion | Ask 3 different models, get one merged answer |
Rate limits, daily quotas, server errors, timeouts, empty answers, providers that go silent: it moves on before your app sees anything.
Measures each model's real speed and reliability, reads Retry-After and rate-limit headers, rotates several keys per provider.
OpenAI Chat & Responses (Codex), Anthropic Messages (Claude Code), Ollama, embeddings and an MCP server.
Rescues tool calls written as text, hides <think> blocks, retries without unsupported parameters, handles provider quirks.
Providers and models come live from free-llm.com and refresh on their own. No paid plan to get updates faster.
Runs on your computer. Keys are AES-256-GCM encrypted and only sent to their own provider. No account, no telemetry.
auto:fusion asks three different free models and merges the best answer — handy to double-check important work.
About 100 KB, zero dependencies, Node.js 18+. Dashboard, CLI, background service, Docker.
| One free key by hand | A paid AI gateway | Free-LLM Router | |
|---|---|---|---|
| Cost | Free | Pay per token | Free |
| When the free quota runs out | Your tool stops | Keeps going (you pay) | Switches to another free model |
| Free models available | One provider's | Mostly paid | 350+ from 66 providers |
| Claude Code / Codex on free models | Needs a translation proxy | Yes, with paid models | One command |
| New free models | You track them yourself | — | Arrive the same day |
| Where your keys live | Your machine | Their servers | Your machine, encrypted |
A small program that runs on your computer and gives you one address that behaves like the OpenAI or Anthropic API. Behind that address it uses the free plans of many AI providers (Google Gemini, Groq, Mistral, OpenRouter, NVIDIA…) and always sends your request to a free model that is working right now. If one provider is busy or out of quota, it switches to the next one instantly, so your app or coding agent keeps working.
Yes. The router is open source under the MIT license, with no paid tier, no account and no telemetry. The AI models are the providers' own free plans; the router never resells anything and never sees your traffic.
Not to try: guest mode answers right away with a few no-account models (slow and limited). For real work, add 2 or 3 free keys — most providers give one in about a minute without a credit card — and you get dozens of fast, strong models.
Yes. Run "free-llm setup claude" and Claude Code sends its requests to the router, which answers with free models (Opus is mapped to the most capable free models, Sonnet to the best coding models, Haiku to the fastest). Free models are very good but not identical to Claude: for the hardest tasks the quality can be lower.
Anything that speaks the OpenAI or Anthropic API: Claude Code, Codex CLI, OpenCode, Aider, Qwen Code, Goose, Crush, Cursor, Cline, Roo Code, Kilo Code, Continue, Zed, JetBrains AI Assistant, Open WebUI, LibreChat, LobeChat, the OpenAI and Anthropic SDKs, LangChain, LlamaIndex, the Vercel AI SDK and MCP clients.
The list of free providers and models comes live from free-llm.com, where every free tier is checked against official sources. The router refreshes it automatically every few hours, so a new free model reaches every user the same day — no paid plan, no delay.
Only on your computer, in ~/.free-llm, encrypted with AES-256-GCM. Requests go directly from your machine to each provider.
The router only uses your own keys, at your request, like any client app would. Each free plan has its own terms (personal use, rate limits, data use), so read the terms of the providers you sign up for.
One minute to install. Free forever. Open source.
Know a free provider we're missing? Submit it — once checked, it reaches every router the same day.